Latent Planning via Embedding Arithmetic: A Contrastive Approach to Strategic Reasoning

TL;DR

This paper introduces SOLIS, a contrastively learned embedding space for strategic planning in high-dimensional decision tasks like chess, enabling efficient, lightweight planning without explicit policy or dynamics models.

Contribution

The paper presents SOLIS, a novel contrastive learning approach that creates an evaluation-aligned latent space for planning, reducing reliance on traditional models.

Findings

SOLIS achieves competitive chess performance with shallow search.

Evaluation-aligned embeddings facilitate effective planning in high-dimensional spaces.

The approach offers a lightweight alternative to dynamics models or policy learning.

Abstract

Planning in high-dimensional decision spaces is increasingly being studied through the lens of learned representations. Rather than training policies or value heads, we investigate whether planning can be carried out directly in an evaluation-aligned embedding space. We introduce SOLIS, which learns such a space using supervised contrastive learning. In this representation, outcome similarity is captured by proximity, and a single global advantage vector orients the space from losing to winning regions. Candidate actions are then ranked according to their alignment with this direction, reducing planning to vector operations in latent space. We demonstrate this approach in chess, where SOLIS uses only a shallow search guided by the learned embedding to reach competitive strength under constrained conditions. More broadly, our results suggest that evaluation-aligned latent planning offers a lightweight alternative to traditional dynamics models or policy learning.